Medial left-node raising in Japanese Shichi Yatabe University of - PowerPoint PPT Presentation

Medial left-node raising in Japanese Shûichi Yatabe University of Tokyo

Right-node raising (RNR) • Canonical right-node raising This tall and that short student are a couple. (from Shen (2015)) • Medial right-node raising Are you talking about a new or that ex- boyfriend you used to date? (from Chaves (2014))

In this paper, I'm going to show ... • that there is a phenomenon that can be viewed as a mirror image of medial RNR and thus might be designated as medial left- node raising , and • that the properties of this phenomenon are consistent with the predictions of the HPSG- based theory of non-constituent coordination proposed in Yatabe (2001, 2012)

Why this is significant Medial right-node raising is often slightly awkward, and the following two views are both plausible. • A grammar-based view: Medial right-node raising is grammatical. Its slight awkwardness comes from the degraded parallelism between conjuncts. • A performance-based view: Medial right-node raising is a result of a performance error. The grammar-based view predicts the existence of medial left-node raising, a prediction yet to be tested.

Left-node raising (LNR) in English? We went to Paris yesterday and London today. (Probably left-node raising, but hard to distinguish from gapping)

Left-node raising in Japanese This can be elided omoi 'thought' + das- 'to exude' = omoidas- 'to recall'

Questionnaire 1A • 16 respondents • 3 experimental sentences, 29 fillers (for the purpose of this paper). Order of sentences randomized for each respondent. • 4-point scale 1: Perfect 2: Slightly unnatural 3: Considerably unnatural 4: Impossible

A questionnaire result "omoi-" is elided here "impossible" "considerably unnatural" "perfect" "slightly unnatural"

A part of a compound is normally not elidable

The mark in front of each sentence is determined by the average rating M of the sentence, according to the following rule. No mark , when 1 ≦ M < 2 ? , when 2 ≦ M < 2.5 ?? , when 2.5 ≦ M < 3 ?* , when 3 ≦ M < 3.5 * , when 3.5 ≦ M ≦ 4 (Recall that 1 means "perfect", 2 means "slightly unnatural, 3 means "considerably unnatural", and 4 means "impossible")

Questionnaire 1B • 19 respondents • 6 experimental sentences, 37 fillers (for the purpose of this paper). Order of sentences randomized for each respondent. • The same 4-point scale • atar- 'to bump' + chiras 'to sprinkle' = atarichiras- 'to throw tantrums' • tabe- 'to eat' + kir- 'to cut' = tabekir- 'to eat up'

Another example of LNR The boldfaced string is elided here

Yet another example of LNR The boldfaced string is elided here

Non-elidability of the first part of the compound in the second example

Non-elidability of the first part of the compound in the third example

Summary of questionnaires 1A and 1B • Japanese allows left-node raising of part of a compound

If LNR is a mirror image of RNR, then ... • Medial left-node raising must be possible. In other words, it must be possible for the left- node-raised string to be at a non-initial position within the initial conjunct. (Cf. a new _ or that ex- boyfriend you used to date) • It must be impossible for the left-node-raised string to be missing from a non-initial position within a non-initial conjunct. (Cf. *that tall _ you used to date or a new boyfriend )

Questionnaire 2A • 28 respondents • 2 experimental sentences,14 fillers (for the purpose of this paper). Order of sentences randomized for each respondent. • The same 4-point scale • omoi 'thought' + das- 'to exude' = omoidas- 'to recall'

The LNRed expression can be realized at a non-initial position in the initial conjunct "omoi" is elided here

LNR is not possible from a non- initial position in the final conjunct "omoi" is elided here

The subtlety of the contrast was part of the prediction • Medial left-node raising was expected to be slightly awkward, just like medial right- node raising. • The example showing that the first part of the compound omoidas- is normally not elidable was in the "??" range, so the example of impossible left-node raising was predicted to be in the "??" range, too.

Questionnaire 2B • 27 respondents • 4 experimental sentences, 12 fillers (for the purpose of this paper). Order of sentences randomized for each respondent. • The same 4-point scale • atar- 'to bump' + chiras 'to sprinkle' = atarichiras- 'to throw tantrums' • tabe- 'to eat' + kir- 'to cut' = tabekir- 'to eat up'

The second example of medial LNR The two boldfaced strings are elided here

The second example that shows that LNR is not possible from a non-initial position in the final conjunct The boldfaced string is elided here

The third example of medial LNR The two boldfaced strings are elided here

The third example that shows that LNR is not possible from a non-initial position in the final conjunct The boldfaced string is elided here

Summary of Questionnaires 2A and 2B • Medial left-node raising is possible. In other words, it is possible for the left- node-raised string to be at a non-initial position within the initial conjunct. • It is impossible for the left-node-raised string to be missing from a non-initial position within a non-initial conjunct.

A linearization-based account • Each node in a syntactic tree is associated with an order domain, which is a list of domain objects, which are essentially prosodic constituents that are semantically interpreted. • Right-node raising and left-node raising take place in order domains. (Yatabe 2001, 2012) • RNR and LNR come in 2 types: a phonological type, which is merely prosodic ellipsis, and a syntactic type, which involves merging of domain objects.

two domain objects merged

two domain objects merged one string deleted

The persistence constraint • Any ordering relation that holds between domain objects α and β in one order domain must also hold between α and β in all other order domains that α and β are members of. (Kathol 1995) • The generalized persistence constraint: (substitute "strings" for "domain objects" in the above)

Right-node raising and the persistence constraint this tall student (and) that short student → This tall and that short student (Blue>Black and Red>Black throughout) a new boyfriend (or) that ex- boyfriend you used to date → a new or that ex- boyfriend you used to date (Blue>Black, Red>Black, and Black>Green throughout)

Right-node raising and the persistence constraint (continued) that tall boyfriend you used to date (or) a new boyfriend → *that tall you used to date or a new boyfriend (Black>Green at first, but Green>Black in the new structure)

Left-node raising and the persistence constraint pre-LNR sô yû toki ni sukoshi wa atari chirasu no ka sô yû toki ni atari chirasanai no ka ↓ post-LNR sô yû toki ni sukoshi wa atari chirasu no ka chirasanai no ka Blue>Green>Black>Red and Blue>Black>Gold throughout

Left-node raising and the persistence constraint (Continued) pre-LNR sô yû toki ni atari chirasu no ka sô yû toki ni sukoshi mo atari chirasanai no ka ↓ post-LNR * sô yû toki ni atari chirasu no ka sukoshi mo chirasanai no ka Green>Black at first but Black>Green in the new structure

Comparison with Categorial Grammar-based accounts • Categorial Grammar-based account cannot capture the medial LNR facts. • The linearization-based accounts can explain the semantic effects of RNR and LNR, pace Kubota and Levine (2015).

Semantic interpretation in the linearization-based account • In most theories, larger and larger syntactic constituents are interpreted, on the basis of the meaning of smaller syntactic constituents. • In contrast, in the theory of Yatabe (2001), larger and larger domain objects are interpreted, mainly on the basis of the meaning of smaller domain objects. • Therefore, in the latter theory, when two or more domain objects are merged, the semantic interpretation is naturally affected.

Semantic interpretation in the standard theory

Semantic interpretation in the proposed theory

Comparison with SLASH- based accounts • Meaning-preserving RNR = ellipsis and Meaning-changing RNR = movement? • No. • Carl Philip Emmanuel Bach secretly hid or donated every manuscript in his father's collection to the library. (Many of the former type remain lost, while the latter are well pre-served.) (from Warstadt (2015))

Summary • There is a phenomenon that can be viewed as a mirror image of medial RNR and thus might be designated as medial left-node raising. • The properties of this phenomenon are mostly consistent with the predictions of the HPSG-based theory of non-constituent coordination proposed in Yatabe (2001, 2012).